AI Agents Could Drive 10,000x More Web Activity | Former Twitter CEO
Oct 24, 2026 · 59:17
Parallel Web Systems founder and CEO Parag Agrawal, Twitter's former engineer, CTO and CEO, argues agents will use the web 5,000–10,000x more than humans ever have — his original 1,000x prediction was too low, and agents today are only at 10–20% of human usage. He explains how Parallel builds GPU-accelerated search infrastructure rented alongside inference (one to two GPUs for search per ten spent on model inference), counting customers like Harvey and Notion among early demanders of quality, sub-200ms latency, and token efficiency. Because agents don't see ads or click pages, he argues ads and per-seat subscriptions break, so Parallel is building an 'AdSense'-style system that pays content owners when agents read their content, keeping the web open. He also covers GPU fire drills sourced partly over iMessage, growing the team from 30 to 120, why the web will shift from pull to push, why pressure is 'made up' but feeling insecure drives him, and how his wife Benita pushed his ambition higher.
